Sapphirine/201912-20-Music-Emotion-Recognition
Music is a medium to express emotion. According to literature, music emotion can be quantified continuously as valence and arousal (VA) density distribution on a 2-D space. However, these data are hard to retrieve as they require intense human effort to manually label songs, especially the number of songs become enormous. The goal of this project is to reproduce a model proposed by Chin, Y.-H. et.al (2018), to predict VA density for a new song based on those densities for training songs as well as audio features of both new and training songs. This will help save human labeling effort on new songs in the future. Furthermore, a prototype of content-based music recommender system is built to demonstrate the usability of the algorithm.
正在读取项目信息…